AI-Driven Detection and Staging of Acute Lymphoblastic Leukemia from Gene Expression and Morphological Characteristics

Acute Lymphoblastic Leukemia (ALL) is a rapidly progressing type of blood cancer, and early diagnosis plays a key role in improving patient care. This work introduces a twophase approach that combines data from gene expression profiles and blood smear images to support both detection and staging of ALL. In the first phase, gene data is analyzed using Random Forest and Logistic Regression models to identify whether a sample is ALL-positive. The second phase focuses on classifying disease progression by analyzing microscope images into four stages: Benign, Early, Pre-leukemic, and Progressive. This is achieved through three strategies: a custom Convolutional Neural Network (CNN), a fine-tuned ResNet-50 model, and a hybrid model where deep features from ResNet are used with a Support Vector Machine (SVM). Among these, the hybrid method showed the most consistent performance. The integration of molecular and image data provides a reliable tool that may help enhance ALL diagnosis in clinical settings.